THE STABILITY OF LINEAR PREDICTIVE DESCRIPTIONS OF CURVES USED IN HANDWRITING

Abstract

The technique of signal analysis by adaptive predictive filtering has previously been extended to the case of multi-dimensional signals, and it has been conjectured that this might be used to analyze handwriting as a cycloidal curve which is modulated to form the letter shapes. Although a second-order two-dimensional filter can model cycloids, the analysis procedure becomes unstable, and we show how differencing the input and applying a first-order filter rectifies this problem.